netdev - [PATCH nf-next v4 1/5] net: sched: Micro-optimize egress handling

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Fri, 22 Jan 2021 09:47:01 +0100
From:   Lukas Wunner <lukas@...ner.de>
To:     "Pablo Neira Ayuso" <pablo@...filter.org>,
        Jozsef Kadlecsik <kadlec@...filter.org>,
        Florian Westphal <fw@...len.de>
Cc:     netfilter-devel@...r.kernel.org, coreteam@...filter.org,
        netdev@...r.kernel.org, Daniel Borkmann <daniel@...earbox.net>,
        Alexei Starovoitov <ast@...nel.org>,
        Eric Dumazet <edumazet@...gle.com>,
        Thomas Graf <tgraf@...g.ch>,
        Laura Garcia Liebana <nevola@...il.com>,
        John Fastabend <john.fastabend@...il.com>
Subject: [PATCH nf-next v4 1/5] net: sched: Micro-optimize egress handling

sch_handle_egress() returns either the skb or NULL to signal to its
caller __dev_queue_xmit() whether a packet should continue to be
processed.

The skb is always non-NULL, otherwise __dev_queue_xmit() would hit a
NULL pointer deref right at its top.

But the compiler doesn't know that.  So if sch_handle_egress() signals
success by returning the skb, the "if (!skb) goto out;" statement
results in a gratuitous NULL pointer check in the Assembler output.

Avoid by telling the compiler that __dev_queue_xmit() is never passed a
NULL skb.  This also eliminates another gratuitous NULL pointer check in
__dev_queue_xmit()
  qdisc_pkt_len_init()
    skb_header_pointer()
      __skb_header_pointer()

The speedup is barely measurable:
Before: 1877 1875 1878 1874 1882 1873 Mb/sec
After:  1877 1877 1880 1883 1888 1886 Mb/sec

However we're about to add a netfilter egress hook to __dev_queue_xmit()
and without the micro-optimization, it will result in a performance
degradation which is indeed measurable:
With netfilter hook:               1853 1852 1850 1848 1849 1851 Mb/sec
With netfilter hook + micro-optim: 1874 1877 1881 1875 1876 1876 Mb/sec

The performance degradation is caused by a JNE instruction ("if (skb)")
being flipped to a JE instruction ("if (!skb)") once the netfilter hook
is added.  The micro-optimization removes the test and jump instructions
altogether.

Measurements were performed on a Core i7-3615QM.  Reproducer:
ip link add dev foo type dummy
ip link set dev foo up
tc qdisc add dev foo clsact
tc filter add dev foo egress bpf da bytecode '1,6 0 0 0,'
modprobe pktgen
echo "add_device foo" > /proc/net/pktgen/kpktgend_3
samples/pktgen/pktgen_bench_xmit_mode_queue_xmit.sh -i foo -n 400000000 -m "11:11:11:11:11:11" -d 1.1.1.1

Signed-off-by: Lukas Wunner <lukas@...ner.de>
Cc: John Fastabend <john.fastabend@...il.com>
Cc: Daniel Borkmann <daniel@...earbox.net>
Cc: Alexei Starovoitov <ast@...nel.org>
Cc: Eric Dumazet <edumazet@...gle.com>
Cc: Thomas Graf <tgraf@...g.ch>
---
 net/core/dev.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/net/core/dev.c b/net/core/dev.c
index 7afbb642e203..4c16b9932823 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4072,6 +4072,7 @@ struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
  *      the BH enable code must have IRQs enabled so that it will not deadlock.
  *          --BLG
  */
+__attribute__((nonnull(1)))
 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
 {
 	struct net_device *dev = skb->dev;
-- 
2.29.2